Vision and Mind: Chap15 - Ways of Coloring: Comparative ...

This excerpt from

Language Form and Language Function. Frederick J. Newmeyer. ? 1998 The MIT Press.

is provided in screen-viewable form for personal use only by members of MIT CogNet.

Unauthorized use or dissemination of this information is expressly forbidden.

If you have any questions about this material, please contact cognetadmin@cognet.mit.edu.

15

Ways of Coloring: Comparative Color Vision as a Case Study for Cognitive Science

Evan Thompson, Adrian Palacios, and Francisco J. Varela

The study of color vision provides a microcosm of research in cognitive science: Each of the disciplines that compose cognitive science has made significant contributions to our understanding of color. Neuroscientists have had some success in uncovering the anatomical and physiological correlates of color vision in the visual system, primarily in primates (DeValois and DeValois 1975; Livingstone and Hubel 1984; Zeki 1983); cellular biologists have characterized the retinal basis of sensitivity (Brown and Wald 1964; Dartnall et al. 1983; Svaetichin and MacNichol 1958); molecular biologists have isolated and sequenced the genes for the three different types of color-sensitive photopigments in the human eye (Nathans et al. 1986); psychophysicists have contributed quantitative models for human visual performance (Hurvich 1985; Hurvich and Jameson 1957; Jameson 1985); cognitive psychologists have provided models of the structure of human color categories (Heider 1972; Rosch 1973); linguists have shown that human languages contain a limited number of "basic" color terms (Berlin and Kay 1969) and have provided models to derive these semantic universals from properties of the visual system (Kay and McDaniel 1978); researchers in computational vision and artificial intelligence have devised computational models and algorithms for color constancy (Gershon 1987; Hurlbert 1986; Land 1983; Maloney 1985; Maloney and Wandell 1986); and finally, philosophers have discussed the ontological status of color and its implications for theories of mind and perception (Hardin 1988; Hilbert 1987; Thompson 1989).

This target article is intended as a contribution to this ongoing interdisciplinary effort. We propose to offer here a new empirical and philosophical perspective on color vision, one based on recent experimental research in comparative color vision--studies of color vision in various animal species. We do not intend to provide a detailed scientific review of current research on this topic (see Goldsmith

352 Evan Thompson, Adrian Palacios, and Francisco J. Varela

1990; Jacobs 1981; and Nuboer 1986). Rather, we wish to draw on this material, especially recent research on fishes, birds, and insects, to cast new light on some fundamental questions in visual perception, cognitive science, and the philosophy of mind.

Our presentation has three stages. In the first, we provide an overview of various types of explanation for color vision in contemporary visual science, showing how particular types of explanation have been used to motivate various views about what color is, that is, about the ontology of color. As we shall see, those who favor objectivism about color, the view that colors are perceiver-independent physical properties (Hilbert 1987; Matthen 1988), rely on computational vision, whereas those who favor subjectivism, the view that colors are internal sensory qualities (Hardin 1988), rely on psychophysics and neurophysiology. In the second stage, we propose a broader comparative and evolutionary perspective on color vision. We present what we call "the comparative argument," which purports to show that an adequate account of color must be experientialist (unlike objectivism) and ecological (unlike subjectivism). In the third stage, we explore the implications of the comparative argument for vision research. We argue that the typical emphasis in computational vision on optimally "recovering" prespecified features of the environment (i.e., distal properties whose specification is thought to be independent of the sensory-motor capacities of the animal) is unsatisfactory. Instead, visual perception is better conceived as the visual guidance of activity in an environment that is constituted largely by that very activity. Here we present what we call an "enactive" approach to perception (proposed originally by Varela 1979; 1984; 1989; 1991a; and developed subsequently by Varela et al. 1991). We then suggest some directions for further research that follow from our discussion.

1 Explanation in Visual Science and the Ontology of Color

1.1 Levels of Explanation: A Brief Overview A central concern in contemporary visual science (indeed throughout all cognitive science) is the relation among various levels of generalization and explanation. Following Churchland and Sejnowski (1988), we can distinguish several notions of "level" at work in cognitive science: levels of analysis, of organization, and of operation ("processing"). Because these notions will prove to be of use in our discussion of color vision, we review them briefly here.

In vision research, the notion of levels of analysis is most familiar from the work of Marr and Poggio (1977). In their framework, vision requires analysis and expla-

Ways of Coloring 353

nation at three different levels: (i) the level of computational theory; (ii) the level of algorithm; and (iii) the level of physical implementation. The computational level is an abstract analysis of the problem or task, which for early vision, according to Marr and Poggio, is the recovery of three-dimensional scenes from ambiguous twodimensional projections, otherwise known as "inverse optics" (Marr 1982; Poggio et al. 1985). For color vision, the inverse optics problem is to recover the invariant surface spectral reflectances of objects in a scene. The algorithmic level is concerned with the specific formal procedures required to perform a given computational task. Finally, the level of physical implementation is concerned with how the algorithms are physically realized in biological or artificial systems.

It is well known that Marr (1982) claimed that these three levels of analysis were largely independent. In the study of biological vision, Marr also supposed that the algorithmic level corresponds to psychophysics and to parts of neurophysiology, whereas the implementational level corresponds to most of neurophysiology and neuroanatomy (1982, p. 26). This conception of explanation in visual science, especially as applied to the study of natural vision, has generated considerable discussion and debate. Among other things, many dispute Marr's (1982) claim that the three levels of analysis are largely independent. Some favor a more "bottom up" approach to the explanation of visual processes, and some criticize Marr's assumption of optimality at the computational level, that is, that "what is being computed is optimal in some sense or is guaranteed to function correctly" (1982, p. 19) [see also Schoemaker, "The Quest for Optimality: A Positive Heuristic of Science?" BBS 14(2) 1991; and Anderson, "Is Human Cognition Adaptive?" BBS 14(3) 1991.] We do not intend to review all of these controversies here.1 We mention them, rather, as pointers toward some of the issues that will arise shortly when we discuss models of color vision, and when we present our alternative "enactive" approach to visual perception in section 3.

In contrast to the notion of levels of analysis, the notion of levels of organization is relatively straightforward. In the nervous system, we find highly organized structures at many different scales from molecules to synapses, neurons, neuronal ensembles, neural networks, maps, systems, and so on. Each level has properties specific to it, which in turn require different techniques for their investigation. Such organizational complexity is certainly evident in color vision, ranging from the chemical properties of receptor photopigments to the network properties of retinal and cortical cells.

Finally, in addition to these levels of organization, we find many levels of operation in the nervous system. How these levels are to be assigned, however, is con-

354 Evan Thompson, Adrian Palacios, and Francisco J. Varela

siderably less clear than it is for levels of organization. The typical procedure is to order the levels hierarchically from peripheral (lower) to central (higher) areas (measured in terms of synaptic distance from sensory stimulation), thereby suggesting that "processing" in the nervous system proceeds sequentially. We wish, however, to dissociate the notion of levels of operation from the idea that processing among the levels is sequential. If (as we and many others believe) "higher" levels can significantly affect the processing in "lower" levels, then the notion of sequential processing will be of limited application, or at least will have to be modified considerably. To cite just one example that is relevant for our discussion here: Although the visual system is typically described as carrying out sequential processing from retina to lateral geniculate nucleus (LGN) to visual cortex, it is also well known that there are massive back-projections from all areas of the cortex to the thalamic nuclei (Steriade and Deschenes 1985). In the case of the visual system, there are actually more fibers going down from the visual cortex to the LGN than go in the reverse direction (Robson 1983). This organization suggests that neuronal activity in central levels may considerably modulate the activity at peripheral levels, an idea that is also supported by some recent experiments (e.g., Varela and Singer 1987). We set this issue aside here. However the relations among levels of operation must ultimately be conceptualized, it is obvious that there are various levels to be distinguished. For example, in primate color vision, we need to understand at the very least the two-way interactions between operations in the retina, thalamus, striate (VI) and peristriate (V4) visual cortex.

With these three notions of "level" in hand we can now turn specifically to color vision. In the remainder of section 1 we give a brief overview of the types of explanation offered for color vision, showing how they have been used to motivate contrasting philosophical positions on the ontology of color.

1.2 Color Space: Psychophysics and Neurophysiology In general, psychophysics and neurophysiology have taken as their point of departure what is known as "color space." This is the closed space formed by the three semi-independent dimensions of color known as hue, chroma or saturation, and value or brightness (figure 15.1).2 Hue obviously refers to the redness, greenness, yellowness, or blueness of a given color. Saturation refers to the proportion of hue in a given color relative to the achromatic (white-black) dimension: Saturated colors have a comparatively greater degree of hue, whereas desaturated colors are comparatively closer to gray. Brightness refers to the achromatic or white-black dimen-

 ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download